Memory’s Next Move, Google’s Fork in the Road, Countdown to July Jobs Report 8/6/26

**SPEAKER_1** (0:00)
Did you know that NeuroSymbolic AI can help your business find new ways to make money? EY Parthenon is the only one offering this groundbreaking growth platform. NeuroSymbolic AI analyzes hundreds of millions of data points to reveal new growth strategies. EY Parthenon teams deploy this innovation so you can uncover hidden value and scale beyond existing boundaries. Want to know what NeuroSymbolic AI can do for your business? Contact EY Parthenon today. Solutions that work in practice, not just on paper.

**SPEAKER_2** (0:30)
This episode is brought to you by Schwab Market Update, an original podcast from Charles Schwab. Join host Keith Lansford for this information-packed daily market preview delivered in 10 minutes or less, including projected stock updates, monetary policy decisions, and key results and statistics that may impact your trading. Download the latest episode and subscribe at schwab.com/marketupdatepodcast, or find Schwab Market Update wherever you get your podcasts.

**Kelly Evans** (0:58)
You're listening to The Exchange. Here's today's show.
Thank you very much, Scott. Some big earnings decliners are weighing on the markets today. I'm Kelly Evans, and welcome to The Exchange. AppLovin, Datadog, and Honeywell Aerospace, all dropping sharply as you can see. They're nearly 20% on the session, as the strong earnings momentum we've seen so far does hit a soft patch. Our guest says this shakeout makes value investing more powerful than ever. Wait, hear what he has to say about it. Matt Miskin is here to explain. Plus, has the memory trade peaked? Sandisk and Western Digital seeing huge declines after their earnings last night, but Sandisk is only down about 4% right now, 10 for Western Digital. As a new report says, NVIDIA is weighing a radical idea, using less memory in its next-gen chips.
And the turnover at Google revealing a larger struggle inside the company, whether to rent out its compute or use it for its own research purposes. Mackenzie Segalos will have more on that. But let's begin with this whipsaw action in the memory names. After falling 10 last night, Sandisk is still down about 4%, Western Digital's comeback not as big, but it was down 16% earlier on. Its earnings didn't miss, the outlook didn't miss, but the whispered numbers were a hurdle neither one could clear. Let's bring in Susquehanna's Mehdi Hosseini for today's opening exchange. Mehdi, it's great to see you again, and what does the market reaction tell you?

**Mehdi Hosseini** (2:20)
Well, thanks for having me on the program. I think it was a week or so ago that I was in your program. We were talking about when is a good time to buy memory. And I think we talked about the late summer will be the right time. And this is how I'm answering your question. I think we're going through this process of normalization where margins are peaking, but we don't know where margins would go looking forward.
All of these memory companies are also talking about a long term contract. We really don't know the economics of the contract. So it's going to take us some time to better figure out the normalized margin, how earnings are going to be like looking into next year. And I think that would set up a late summer, perhaps would offer a better setup for memory names.

**Kelly Evans** (3:07)
Well, I'll come back to that in a second. Information is out with a new report just this afternoon. Mehdi, I don't know if you've seen it yet, but they're saying that... Yeah, well, I'll tell the audience who maybe hasn't seen it yet, but they're reporting that Nvidia is thinking about using less memory for its next generation Rubin chips. What are your thoughts?

**Mehdi Hosseini** (3:27)
I'm not sure if I would agree with that. I think the next platform, Rubin Ultra, is got some technical issues. I think the data rate, the networking wall is a bigger issue.
Yes, memory is costing more, but memory wall is also very important. I think those are sort of a behavior you see from key customers because they have to deal with how much memory cost has increased. But it doesn't really change the fact that as you look into the next generation of AI platforms, it requires more memory, networking becomes an issue. And on top of that, we're transitioning from training to inferencing. And all of these factors would give us some time, would require us some time to better understand the dynamics. But no, I'm not so much concerned about de-especting if anything memory content is going higher. We just don't know how the margins and normalized margins would look like.

**Kelly Evans** (4:32)
But do you think there's an Nvidia problem with the, what is it called, Rubin? Ultra, what is it called?